1 /*
2 * Copyright(c) 2011-2016 Intel Corporation. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 * Eddie Dong <eddie.dong@intel.com>
25 * Kevin Tian <kevin.tian@intel.com>
26 *
27 * Contributors:
28 * Zhi Wang <zhi.a.wang@intel.com>
29 * Changbin Du <changbin.du@intel.com>
30 * Zhenyu Wang <zhenyuw@linux.intel.com>
31 * Tina Zhang <tina.zhang@intel.com>
32 * Bing Niu <bing.niu@intel.com>
33 *
34 */
35
36 #include "i915_drv.h"
37 #include "gvt.h"
38 #include "trace.h"
39
40 struct render_mmio {
41 int ring_id;
42 i915_reg_t reg;
43 u32 mask;
44 bool in_context;
45 u32 value;
46 };
47
48 static struct render_mmio gen8_render_mmio_list[] __cacheline_aligned = {
49 {RCS, _MMIO(0x229c), 0xffff, false},
50 {RCS, _MMIO(0x2248), 0x0, false},
51 {RCS, _MMIO(0x2098), 0x0, false},
52 {RCS, _MMIO(0x20c0), 0xffff, true},
53 {RCS, _MMIO(0x24d0), 0, false},
54 {RCS, _MMIO(0x24d4), 0, false},
55 {RCS, _MMIO(0x24d8), 0, false},
56 {RCS, _MMIO(0x24dc), 0, false},
57 {RCS, _MMIO(0x24e0), 0, false},
58 {RCS, _MMIO(0x24e4), 0, false},
59 {RCS, _MMIO(0x24e8), 0, false},
60 {RCS, _MMIO(0x24ec), 0, false},
61 {RCS, _MMIO(0x24f0), 0, false},
62 {RCS, _MMIO(0x24f4), 0, false},
63 {RCS, _MMIO(0x24f8), 0, false},
64 {RCS, _MMIO(0x24fc), 0, false},
65 {RCS, _MMIO(0x7004), 0xffff, true},
66 {RCS, _MMIO(0x7008), 0xffff, true},
67 {RCS, _MMIO(0x7000), 0xffff, true},
68 {RCS, _MMIO(0x7010), 0xffff, true},
69 {RCS, _MMIO(0x7300), 0xffff, true},
70 {RCS, _MMIO(0x83a4), 0xffff, true},
71
72 {BCS, _MMIO(0x2229c), 0xffff, false},
73 {BCS, _MMIO(0x2209c), 0xffff, false},
74 {BCS, _MMIO(0x220c0), 0xffff, false},
75 {BCS, _MMIO(0x22098), 0x0, false},
76 {BCS, _MMIO(0x22028), 0x0, false},
77 };
78
79 static struct render_mmio gen9_render_mmio_list[] __cacheline_aligned = {
80 {RCS, _MMIO(0x229c), 0xffff, false},
81 {RCS, _MMIO(0x2248), 0x0, false},
82 {RCS, _MMIO(0x2098), 0x0, false},
83 {RCS, _MMIO(0x20c0), 0xffff, true},
84 {RCS, _MMIO(0x24d0), 0, false},
85 {RCS, _MMIO(0x24d4), 0, false},
86 {RCS, _MMIO(0x24d8), 0, false},
87 {RCS, _MMIO(0x24dc), 0, false},
88 {RCS, _MMIO(0x24e0), 0, false},
89 {RCS, _MMIO(0x24e4), 0, false},
90 {RCS, _MMIO(0x24e8), 0, false},
91 {RCS, _MMIO(0x24ec), 0, false},
92 {RCS, _MMIO(0x24f0), 0, false},
93 {RCS, _MMIO(0x24f4), 0, false},
94 {RCS, _MMIO(0x24f8), 0, false},
95 {RCS, _MMIO(0x24fc), 0, false},
96 {RCS, _MMIO(0x7004), 0xffff, true},
97 {RCS, _MMIO(0x7008), 0xffff, true},
98 {RCS, _MMIO(0x7000), 0xffff, true},
99 {RCS, _MMIO(0x7010), 0xffff, true},
100 {RCS, _MMIO(0x7300), 0xffff, true},
101 {RCS, _MMIO(0x83a4), 0xffff, true},
102
103 {RCS, _MMIO(0x40e0), 0, false},
104 {RCS, _MMIO(0x40e4), 0, false},
105 {RCS, _MMIO(0x2580), 0xffff, true},
106 {RCS, _MMIO(0x7014), 0xffff, true},
107 {RCS, _MMIO(0x20ec), 0xffff, false},
108 {RCS, _MMIO(0xb118), 0, false},
109 {RCS, _MMIO(0xe100), 0xffff, true},
110 {RCS, _MMIO(0xe180), 0xffff, true},
111 {RCS, _MMIO(0xe184), 0xffff, true},
112 {RCS, _MMIO(0xe188), 0xffff, true},
113 {RCS, _MMIO(0xe194), 0xffff, true},
114 {RCS, _MMIO(0x4de0), 0, false},
115 {RCS, _MMIO(0x4de4), 0, false},
116 {RCS, _MMIO(0x4de8), 0, false},
117 {RCS, _MMIO(0x4dec), 0, false},
118 {RCS, _MMIO(0x4df0), 0, false},
119 {RCS, _MMIO(0x4df4), 0, false},
120
121 {BCS, _MMIO(0x2229c), 0xffff, false},
122 {BCS, _MMIO(0x2209c), 0xffff, false},
123 {BCS, _MMIO(0x220c0), 0xffff, false},
124 {BCS, _MMIO(0x22098), 0x0, false},
125 {BCS, _MMIO(0x22028), 0x0, false},
126
127 {VCS2, _MMIO(0x1c028), 0xffff, false},
128
129 {VECS, _MMIO(0x1a028), 0xffff, false},
130
131 {RCS, _MMIO(0x7304), 0xffff, true},
132 {RCS, _MMIO(0x2248), 0x0, false},
133 {RCS, _MMIO(0x940c), 0x0, false},
134 {RCS, _MMIO(0x4ab8), 0x0, false},
135
136 {RCS, _MMIO(0x4ab0), 0x0, false},
137 {RCS, _MMIO(0x20d4), 0x0, false},
138
139 {RCS, _MMIO(0xb004), 0x0, false},
140 {RCS, _MMIO(0x20a0), 0x0, false},
141 {RCS, _MMIO(0x20e4), 0xffff, false},
142 };
143
144 static u32 gen9_render_mocs[I915_NUM_ENGINES][64];
145 static u32 gen9_render_mocs_L3[32];
146
handle_tlb_pending_event(struct intel_vgpu * vgpu,int ring_id)147 static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
148 {
149 struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
150 enum forcewake_domains fw;
151 i915_reg_t reg;
152 u32 regs[] = {
153 [RCS] = 0x4260,
154 [VCS] = 0x4264,
155 [VCS2] = 0x4268,
156 [BCS] = 0x426c,
157 [VECS] = 0x4270,
158 };
159
160 if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
161 return;
162
163 if (!test_and_clear_bit(ring_id, (void *)vgpu->tlb_handle_pending))
164 return;
165
166 reg = _MMIO(regs[ring_id]);
167
168 /* WaForceWakeRenderDuringMmioTLBInvalidate:skl
169 * we need to put a forcewake when invalidating RCS TLB caches,
170 * otherwise device can go to RC6 state and interrupt invalidation
171 * process
172 */
173 fw = intel_uncore_forcewake_for_reg(dev_priv, reg,
174 FW_REG_READ | FW_REG_WRITE);
175 if (ring_id == RCS && (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)))
176 fw |= FORCEWAKE_RENDER;
177
178 intel_uncore_forcewake_get(dev_priv, fw);
179
180 I915_WRITE_FW(reg, 0x1);
181
182 if (wait_for_atomic((I915_READ_FW(reg) == 0), 50))
183 gvt_vgpu_err("timeout in invalidate ring (%d) tlb\n", ring_id);
184 else
185 vgpu_vreg(vgpu, regs[ring_id]) = 0;
186
187 intel_uncore_forcewake_put(dev_priv, fw);
188
189 gvt_dbg_core("invalidate TLB for ring %d\n", ring_id);
190 }
191
load_mocs(struct intel_vgpu * vgpu,int ring_id)192 static void load_mocs(struct intel_vgpu *vgpu, int ring_id)
193 {
194 struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
195 i915_reg_t offset, l3_offset;
196 u32 regs[] = {
197 [RCS] = 0xc800,
198 [VCS] = 0xc900,
199 [VCS2] = 0xca00,
200 [BCS] = 0xcc00,
201 [VECS] = 0xcb00,
202 };
203 int i;
204
205 if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
206 return;
207
208 offset.reg = regs[ring_id];
209 for (i = 0; i < 64; i++) {
210 gen9_render_mocs[ring_id][i] = I915_READ_FW(offset);
211 I915_WRITE(offset, vgpu_vreg(vgpu, offset));
212 offset.reg += 4;
213 }
214
215 if (ring_id == RCS) {
216 l3_offset.reg = 0xb020;
217 for (i = 0; i < 32; i++) {
218 gen9_render_mocs_L3[i] = I915_READ_FW(l3_offset);
219 I915_WRITE_FW(l3_offset, vgpu_vreg(vgpu, l3_offset));
220 l3_offset.reg += 4;
221 }
222 }
223 }
224
restore_mocs(struct intel_vgpu * vgpu,int ring_id)225 static void restore_mocs(struct intel_vgpu *vgpu, int ring_id)
226 {
227 struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
228 i915_reg_t offset, l3_offset;
229 u32 regs[] = {
230 [RCS] = 0xc800,
231 [VCS] = 0xc900,
232 [VCS2] = 0xca00,
233 [BCS] = 0xcc00,
234 [VECS] = 0xcb00,
235 };
236 int i;
237
238 if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
239 return;
240
241 offset.reg = regs[ring_id];
242 for (i = 0; i < 64; i++) {
243 vgpu_vreg(vgpu, offset) = I915_READ_FW(offset);
244 I915_WRITE_FW(offset, gen9_render_mocs[ring_id][i]);
245 offset.reg += 4;
246 }
247
248 if (ring_id == RCS) {
249 l3_offset.reg = 0xb020;
250 for (i = 0; i < 32; i++) {
251 vgpu_vreg(vgpu, l3_offset) = I915_READ_FW(l3_offset);
252 I915_WRITE_FW(l3_offset, gen9_render_mocs_L3[i]);
253 l3_offset.reg += 4;
254 }
255 }
256 }
257
258 #define CTX_CONTEXT_CONTROL_VAL 0x03
259
260 /* Switch ring mmio values (context) from host to a vgpu. */
switch_mmio_to_vgpu(struct intel_vgpu * vgpu,int ring_id)261 static void switch_mmio_to_vgpu(struct intel_vgpu *vgpu, int ring_id)
262 {
263 struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
264 struct render_mmio *mmio;
265 u32 v;
266 int i, array_size;
267 u32 *reg_state = vgpu->shadow_ctx->engine[ring_id].lrc_reg_state;
268 u32 ctx_ctrl = reg_state[CTX_CONTEXT_CONTROL_VAL];
269 u32 inhibit_mask =
270 _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
271 i915_reg_t last_reg = _MMIO(0);
272
273 if (IS_SKYLAKE(vgpu->gvt->dev_priv)
274 || IS_KABYLAKE(vgpu->gvt->dev_priv)) {
275 mmio = gen9_render_mmio_list;
276 array_size = ARRAY_SIZE(gen9_render_mmio_list);
277 load_mocs(vgpu, ring_id);
278 } else {
279 mmio = gen8_render_mmio_list;
280 array_size = ARRAY_SIZE(gen8_render_mmio_list);
281 }
282
283 for (i = 0; i < array_size; i++, mmio++) {
284 if (mmio->ring_id != ring_id)
285 continue;
286
287 mmio->value = I915_READ_FW(mmio->reg);
288
289 /*
290 * if it is an inhibit context, load in_context mmio
291 * into HW by mmio write. If it is not, skip this mmio
292 * write.
293 */
294 if (mmio->in_context &&
295 ((ctx_ctrl & inhibit_mask) != inhibit_mask) &&
296 i915.enable_execlists)
297 continue;
298
299 if (mmio->mask)
300 v = vgpu_vreg(vgpu, mmio->reg) | (mmio->mask << 16);
301 else
302 v = vgpu_vreg(vgpu, mmio->reg);
303
304 I915_WRITE_FW(mmio->reg, v);
305 last_reg = mmio->reg;
306
307 trace_render_mmio(vgpu->id, "load",
308 i915_mmio_reg_offset(mmio->reg),
309 mmio->value, v);
310 }
311
312 /* Make sure the swiched MMIOs has taken effect. */
313 if (likely(INTEL_GVT_MMIO_OFFSET(last_reg)))
314 I915_READ_FW(last_reg);
315
316 handle_tlb_pending_event(vgpu, ring_id);
317 }
318
319 /* Switch ring mmio values (context) from vgpu to host. */
switch_mmio_to_host(struct intel_vgpu * vgpu,int ring_id)320 static void switch_mmio_to_host(struct intel_vgpu *vgpu, int ring_id)
321 {
322 struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
323 struct render_mmio *mmio;
324 i915_reg_t last_reg = _MMIO(0);
325 u32 v;
326 int i, array_size;
327
328 if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
329 mmio = gen9_render_mmio_list;
330 array_size = ARRAY_SIZE(gen9_render_mmio_list);
331 restore_mocs(vgpu, ring_id);
332 } else {
333 mmio = gen8_render_mmio_list;
334 array_size = ARRAY_SIZE(gen8_render_mmio_list);
335 }
336
337 for (i = 0; i < array_size; i++, mmio++) {
338 if (mmio->ring_id != ring_id)
339 continue;
340
341 vgpu_vreg(vgpu, mmio->reg) = I915_READ_FW(mmio->reg);
342
343 if (mmio->mask) {
344 vgpu_vreg(vgpu, mmio->reg) &= ~(mmio->mask << 16);
345 v = mmio->value | (mmio->mask << 16);
346 } else
347 v = mmio->value;
348
349 if (mmio->in_context)
350 continue;
351
352 I915_WRITE_FW(mmio->reg, v);
353 last_reg = mmio->reg;
354
355 trace_render_mmio(vgpu->id, "restore",
356 i915_mmio_reg_offset(mmio->reg),
357 mmio->value, v);
358 }
359
360 /* Make sure the swiched MMIOs has taken effect. */
361 if (likely(INTEL_GVT_MMIO_OFFSET(last_reg)))
362 I915_READ_FW(last_reg);
363 }
364
365 /**
366 * intel_gvt_switch_render_mmio - switch mmio context of specific engine
367 * @pre: the last vGPU that own the engine
368 * @next: the vGPU to switch to
369 * @ring_id: specify the engine
370 *
371 * If pre is null indicates that host own the engine. If next is null
372 * indicates that we are switching to host workload.
373 */
intel_gvt_switch_mmio(struct intel_vgpu * pre,struct intel_vgpu * next,int ring_id)374 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
375 struct intel_vgpu *next, int ring_id)
376 {
377 struct drm_i915_private *dev_priv;
378
379 if (WARN_ON(!pre && !next))
380 return;
381
382 gvt_dbg_render("switch ring %d from %s to %s\n", ring_id,
383 pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
384
385 dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
386
387 /**
388 * We are using raw mmio access wrapper to improve the
389 * performace for batch mmio read/write, so we need
390 * handle forcewake mannually.
391 */
392 intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
393
394 /**
395 * TODO: Optimize for vGPU to vGPU switch by merging
396 * switch_mmio_to_host() and switch_mmio_to_vgpu().
397 */
398 if (pre)
399 switch_mmio_to_host(pre, ring_id);
400
401 if (next)
402 switch_mmio_to_vgpu(next, ring_id);
403
404 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
405 }
406